Document 14544842

advertisement
The SIJ Transactions on Computer Networks & Communication Engineering (CNCE), Vol. 2, No. 1, January-February 2014
Authentication by Palmprint
Recognition using Phase-Difference
Trained by Probability Neural Network
S.G. Siddharth*, M. Arunkumar** & S. Valarmathi***
*P.G. Scholar, Department of Electronics & Communication Engineering, Bannari Amman Institute of Technology, Sathyamanglam,
Tamilnadu, INDIA. E-Mail: siddharth.bannari{at}gmail{dot}com
**Assistant Professor, Department of Electronics & Communication Engineering, Bannari Amman Institute of Technology, Sathyamanglam,
Tamilnadu, INDIA. E-Mail: mak5phd{at}gmail{dot}com
***Head of the Department, Electronics & Communication Engineering, Bannari Amman Institute of Technology, Sathyamanglam,
Tamilnadu, INDIA.
Abstract—In this paper we are using phase difference method for the purpose of extracting the features and for
training and testing the extracted features Probability Neural Networks is used the extracted features are treated
with PCA. Among the various methods available methods, PNN stands apart from the set because it improves
the effective level of security by just increasing the number of hidden layers instead of changing the
parameters, as in other systems. The basic reason why we use phase difference is that for obtaining high
precision and efficiency (i.e.) we have two parts in phase difference namely the real and imaginary part, to
reduce the system execution time we have chosen only the real part and processed with it. We have various
techniques but we have chosen PCA method as a dimensionality reduction technique since it takes only the
ROI that we have defined before into account. This project aims to provide an effective authentication system.
The combination of PNN with phase difference for sure will be a cost effective, highly secure and a robust
setup. The setup proves to be immune to illumination and other factors that happens to hinder performance
during capture of the input image (palm-print).
Keywords—Authentication; Features; Learning Rate; Palm-Print; Phase Difference; Recognition Rate.
Abbreviations—Back Propagation Neural Network (BPN); Principle Component Analysis (PCA); Probability
Neural Network (PNN); Region of Interest (ROI).
I.
B
INTRODUCTION
IOMETRIC system is essentially a pattern
recognition system which makes a personal
identification by determining the authenticity of a
specific physiological or behavioral characteristic possessed
by the user. Biometrics has gained much attention in the
security world recently. Many different types of personal
identification systems have been developed and palm print
verification is one of the emerging technologies because of its
stable, unique characteristics, low-price capture device, fast
execution speed also it provides a large area for feature
extraction. Palm print recognizes a person based on the
principal lines, wrinkles and ridges on the surface of the
palm. The recognition process consists of image acquisition,
pre-processing, feature extraction, matching and result. There
are many different techniques available for pre-processing,
feature extraction, classifiers [Badrinath & Phalguni Gupta,
2012; Bob Zhang et al., 2013]. Palm print matching is an
approach for verifying a palm print input by matching the
input to the claimed identity template stored in a database. If
ISSN: 2321 – 2403
the dissimilarity measure between the input and the claimed
template is below the predefined threshold value, the palm
print input is verified possessing same identity as the claimed
identity template. Our palm print matching system contains
three modules, preprocessing and ROI Extraction, Feature
extraction, Energy calculation, Voting Classification and
matching.
The main steps of palm print matching system are
described below:
1. Preprocessing: The color image is converted into
gray scale and feature extraction is made at the
central part of the palm print.
2. Feature Extraction: We apply a contour let
transform to extract feature information from the
central part.
3. Energy Calculation: Phase difference is applied to
extract the high dimensional features, which is
enormously present in the palm print. Energy
features are calculated as a feature selection step and
feature vectors are created.
© 2014 | Published by The Standard International Journals (The SIJ)
1
The SIJ Transactions on Computer Networks & Communication Engineering (CNCE), Vol. 2, No. 1, January-February 2014
4.
PNN: An efficient method to test and train the
inputs, validate the palm inputs for effective
authentication is done.
II.
LITERATURE SURVEY
Badrinath & Phalguni Gupta (2012) proposed that the
reconstruction error using principal component analysis, the
feature are extracted by phase difference method and
hamming distance is used to find the matching score.
Bob Zhang et al., (2013) proposed that 3D Palm print has
proved to be a significant biometrics for personnel
authentication. Special methods adapted are coarse level
matching and ranking support vector machine.
Priyanka Somvanshi & Milind Rane (2012) detailed
study about various pattern recognition systems, feature
extraction techniques used and the type of features that are
available.
Dakshin Rajan Kishu et al., (2012) presented that Intramodel fusion environment to integrate multiple raw palm
images at low level. Gabor based feature extraction technique
is being used.
Dewi Yanti Liliana & Eries Tri Utaminingsih (2012)
suggested the methodology for palm recognition consisted of
block-based line detection which dealing with palm print
features and recognizing using Dynamic time wrapping.
Ngoc-Son Vu et al., (2012) examined that the use of
patterns of oriented edge magnitude by applying a selfsimilarity based structure on oriented magnitudes and proves
that it addresses computational cost, robustness and
discriminative power. Results are 20 times better when
compared to gabber filter.
Micheal Choras & Rafal Kozik (2011) proposed
biometrics for contactless and unrestricted access control for
mobile devices. We use knuckles and palm print by fusion.
Lin Zhang et al., (2012) studies image local features
induced by the phase congruency model which is supported
by psychophysical and neurophysiologic evidences and then
the features of palm and knuckle are fused and processed.
Zhenhua Guo et al., (2011) proposed techniques to
improve the palmprint image by providing an accurate light
source and is angle of placement.
David Zhang et al., (2011) proposed that exclusive
extraction of palm print features and palm vein features and
to fuse them together to improve the systems efficiency.
Yangqiang Zhang et al., (2011) presented that single
sample biometrics recognition being at latest trend are prone
to errors so that here the features are fused. The feature
extraction technique used is that locality preserving
projection based on wavelet transform.
Adams Kong et al., (2009) proposed the overview of
current palm print research, describing in particular capture
devices, preprocessing, verification algorithms, algorithms
especially designed for real time palmprint identification and
measures for protecting palmprint systems and users privacy.
ISSN: 2321 – 2403
Tee Connie et al., (2005) focus made on principal
component analysis, fisher discriminates analysis and
independent component analysis for authentication.
Iangqian Wu et al., (2004) proposed that novel method
for classification of low resolution palmprints. The concern
here is the principal lines. Different types of classification
based on principal lines are done and the accuracy is assumed
to be 96.03%.
Jiwen Lu et al., (2008) studied that recognition method
by Locality Preserving Projections and extreme learning
machine neural network. Firstly two dimension discrete
wavelet transform is applied in the region of interest of each
palmprint image for dimensionality reduction, then extreme
learning machine is used to classify the images.
Guangming Lu & David Zhang (2003) proposed that
using Karhunen-loeve transform is used to get the Eigen palm
features.
III.
WORK FLOW
3.1. Feature Extraction using Phase-Difference
Phase difference, rather than amplitude information can
provide the most significant information of an image. Further
it is found to be inherently stable over amplitude and is
independent to intensity levels of the image and hence, the
measurements are invariant to smooth shading and lighting
variations.
One can use phase based features to design an effective
and efficient technique for feature extraction. The phase of 1D-signal x can be obtained using Fourier transform. The
Fourier transform X of time varying 1-D signal x of length M
is given by,
𝑀−1
1 π‘‹π‘š 𝑒 (−2πœ‹π‘– /π‘š )π‘—π‘š
𝑋𝑗 =
(1)
π‘š =0
where j=0,1,……., M-1.
It can be represented in amplitude and phase form as,
π‘₯
(2)
𝑋𝑗 = πœŒπ‘—π‘₯ 𝑒 𝑖∅𝑗
where the magnitude, πœŒπ‘—π‘₯ and the phase of the signal x, ∅𝑗π‘₯ are
given by
(3)
πœŒπ‘—π‘₯ = 𝑅𝑒(𝑋𝑗 )2 + πΌπ‘š((𝑋𝑗 )2
πΌπ‘š 𝑋𝑗
(4)
∅𝑗π‘₯ = (π‘‘π‘Žπ‘›−1
𝑅𝑒 𝑋𝑗
If the X and Y are the Fourier transforms of 1D signals x
and y respectively then the phase difference πœ‘π‘— is obtained by
𝑦
the difference πœ‘π‘—π‘₯ − πœ‘π‘— .
Phase base techniques have successfully applied for a
high accuracy image registration, recognition through iris,
finger print, dental, palm print and other computer based
applications
It is found that the phase difference obtained from the
phase of the signals π‘₯ and 𝑑 is also robust to illumination.
This can be explained as follows. Suppose the illumination of
the two palm print images are different by a constant 𝛼, that
© 2014 | Published by The Standard International Journals (The SIJ)
2
The SIJ Transactions on Computer Networks & Communication Engineering (CNCE), Vol. 2, No. 1, January-February 2014
means, the change in intensity which corresponds to
amplitude can be modeled by scaling the pixel values by 𝛼.
Let g be the intensity scaled 1D signal of x. Then g=𝛼π‘₯
with 𝛼 > 0. The Fourier transform G of g is given by
𝑀−1
π‘”π‘š 𝑒 −2πœ‹π‘–π‘—π‘š /𝑀
𝐺𝑗 =
(5)
π‘š =0
where j=0, 1, …., M-1.
π‘₯
𝐺𝑗 = π›ΌπœŒπ‘—π‘₯ 𝑒 𝑖∅𝑗
(6)
Figure 2: Structure of Probability Neural Network
V.
Figure 1: Flow for Palm Print Feature Extraction
From equation (5) it implies that the phase remains
unaffected due to illumination.
Here for our consideration we have taken the inputs from the
IIT Palm Print database, to which the phase difference is
applied and fed into the BPN. The values tabulated below are
for a fixed set of parameters by varying the spread and
number of persons
Outputs for varying persons and spread.
Table 1: Outputs with Spread Value=0.010
Number of
Elapse
CPU
Persons
Time
Time
2
2.8330
1.965
5
2.5127
2.527
10
2.1967
3.088
15
1.8767
3.650
20
1.5567
4.212
25
1.2359
4.772
3.2. Principle Component Analysis (PCA)
In simple words PCA could be explained to be an existing
system that is used for reducing the number of features to our
required levels. Using a large set of values increases the
system run time and hence an unwanted delay happens and to
avoid it and improve the systems overall efficiency and speed
we are using PCA.
S. No.
1
2
3
4
5
6
PROBABILITY NEURAL NETWORK
A Probabilistic Neural Network (PNN) is predominantly a
classifier
ο‚· Map any input pattern to a number of classifications
ο‚· Can be forced into a more general function
approximate
A PNN is an implementation of a statistical algorithm
called kernel discriminate analysis in which the operations
are organized into a multilayered feed forward network with
four layers.
ο‚· Input layer
ο‚· Pattern layer
ο‚· Summation layer
ο‚· Output lay
The training process of a PNN is essentially the act of
determining the value of the smoothing parameter, sigma
ο‚· Training is fast
ο‚· Determining Sigma
ο‚· Educated guess based on knowledge of the data.
ISSN: 2321 – 2403
ELLAPSE TIME
CPU TIME
Clock
2.824
2.511
2.200
1.890
1.593
1.295
CLOCK
5
4
No. of Persons
IV.
RESULTS AND ANALYSIS
3
2
1
0
1
4
5
Time
Figure 3: Graph for Spread Value = 0.010
S.
No.
1
2
3
4
5
6
2
3
Table 2: Outputs with Spread Value=0.100
Number of
Elapse
CPU
Persons
Time
Time
2
2.313
1.934
5
2.591
2.667
10
2.829
3.400
15
3.148
4.134
20
3.426
4.792
25
3.705
5.525
© 2014 | Published by The Standard International Journals (The SIJ)
6
Clock
2.315
2.589
2.713
2.894
3.168
3.442
3
The SIJ Transactions on Computer Networks & Communication Engineering (CNCE), Vol. 2, No. 1, January-February 2014
ELLAPSE TIME
5
CPU TIME
CLOCK
No. of Persons
4
[2]
[3]
3
2
[4]
1
0
1
2
3
4
5
6
[5]
Time
Figure 4: Graph for Spread Value = 0.010
S. No.
1
2
3
4
5
6
Table 3: Outputs with Spread Value=0.010
Number of
Elapse
CPU
Persons
Time
Time
2
2.131
1.981
5
2.422
2.464
10
2.612
2.948
15
3.003
3.531
20
3.300
4.091
25
3.956
4.575
ELLAPSE TIME
No. of Persons
5
CPU TIME
[6]
Clock
2.13
2.42
2.91
3.01
3.47
4.14
[7]
[8]
[9]
CLOCK
[10]
4
3
[11]
2
1
[12]
0
1
2
3
Time
4
5
6
Figure 5: Graph for Spread Value = 0.010
From the above table it is evident that as the spread value
increases the system acts to be much more speeder.
VI.
CONCLUSION AND FUTURE WORK
The system being designed by using the phase difference
feature extraction method, whose output is fed into the BPN
proves to be highly effective and efficient. The tabulation
even give us an added information that number persons being
in the database also affects the code’s runtime and moreover
the code runs fatly as the spread value increases.
Future enhancement could be that by using principle
component analysis the features could be reduced and KLN
transform be applied to compare it with the existing system.
Using fusion methods systems accuracy level can be
increased. Hardware implementation with the best
combination of the MATLAB codes is done to obtain the
maximum accuracy.
REFERENCES
[1]
[13]
[14]
[15]
[16]
Bob Zhang, Wei Li, Pei Qing & David Zhang (2013), “Palm
Print Classification by Global Features”, IEEE Transactions on
Systems, Man and Cybernetics: Systems, Vol. 43, No. 2.
Priyanka Somvanshi & Milind Rane (2012), “Survey of Palm
Print Recognition”, International Journal of Scientific &
Engineering Research, Vol. 3, No. 2.
Dakshin Rajan Kishu, Ajita Rattan, Phalguni Gupta, Jamuna
Kanta Sing & C. Jinshong Hwang (2012), “Human Identity
Verification using Multispectral Palm Print Fusion”, Journal of
Signal and Information Processing, Vol. 3, Pp. 263–273.
Dewi Yanti Liliana & Eries Tri Utaminingsih (2012), “The
Combination of Palm and Hand Geometry for Biometrics Palm
Recognition”, International Journal for Video and Network
Security, Vol. 2, No. 01.
Ngoc-Son Vu, Hannah M. Dee & Alice Caplier (2012), “Face
Recognition using the POEM Descriptor”, Pattern Recognition,
Vol. 45, Pp. 2478–2488.
Micheal
Choras & Rafal Kozik (2011), “Contactless
Palm Print and Knuckle Biometrics for Mobile Devices”,
Pattern Analysis and Applications, Vol. 15, No. 1, Pp. 73–85.
Lin Zhang, Lei Zhang, David Zhang & Zhenhua Guo (2012),
“Phase Congruency Induced Local Features for FingerKnuckle-Print Recognition”, Pattern Recognition, Vol. 45, Pp.
2522–25311.
Zhenhua Guo, David Zhang, Lei Zhang, Wangmeng Zuo &
Guangming Lu (2011), “Empirical study of light source
selection for Palmprint Recognition”, Pattern Recognition
Letters, Vol. 32, Pp. 120–126.
David Zhang, Zhenhua Guo & Lei Zhang (2011), “Online Joint
Palm Print and Palmvein Verification”, Expert Systems with
Applications, Vol. 38, Pp. 2621–2631.
Yangqiang Zhang, Dongmeni Sun, Zhengding Qiu (2011),
“Hand based Single Sample Biometrics Recognition”, Neural
Computers and Applications, Vol. 21, No. 8, Pp. 835–1844.
Adams Kong, David Zhang & Mohammed Kamel (2009), ”A
Survey of Palm Print Recognition”, Pattern Recognition, Vol.
42, Pp. 1408–1418.
Tee Connie, Andrew Beng Jin & Micheal Goh Kah Ong
(2005), “An Automated Palm Print Recognition system”,
Image and Vision Computing, Vol. 23, Pp. 501–505.
Xiangqian Wu, David Zhang, Kuanquan Wang & Bo Huang
(2004), ”Palmprint Classification using Principal Lines”,
Elsevier, Pattern Recognition, Vol. 37, Pp. 1987–1998..
Jiwen Lu, Yongwei Zhan, Yanxue Xue & Junlin Hu (2008),
“Palm Print Recognition via Locality Preserving Projections
and Extreme Learning Machine Neural Network”,
International Conference on Signal Processing, Pp. 2096–
2099.
Guangming Lu & David Zhang (2003), “Palm Print
Reconstruction using Eigen Palm Features”, Pattern
Recognition Letters, Vol. 23, Pp. 1463–1467.
SG. Siddharth. He received his B.E.,
(Electronics
and
Instrumentation
Engineering) degree from Anna University,
Chennai. He is doing his final year Post
graduate in Embedded Systems (Electronics
and Communication Engineering) in Bannari
Amman
Institute
of
Technology,
Sathyamangalam. He is currently undergoing
his project work in the area of biometrics. His
research interest includes Biometrics, Image Processing and Neural
Networks. He has published 4 papers in International and 6 papers
in National Conferences in various fields.
G.S. Badrinath & Phalguni Gupta (2012), “Palm Print based
Recognition System using Phase – Difference Information”,
Future Generation Computer Systems, Vol. 28, Pp. 287–305.
ISSN: 2321 – 2403
© 2014 | Published by The Standard International Journals (The SIJ)
4
Download